Machine learning models for orthokeratology lens fitting and axial length prediction

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Orthokeratology (OK) is a myopia control method using specialized contact lenses.
• Accurate OK lens fitting and prediction of refractive changes are crucial for effective treatment.
• Existing methods for predicting outcomes can be limited in efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for enhancing orthokeratology lens fitting.
• To predict the axial length of myopic eyes after one year of OK lens wear.
• To improve the overall efficiency and predictive performance in orthokeratology.

Main Methods:

• Retrospective analysis of clinical data from 1302 myopic subjects.
• Implementation of two ML models utilizing demographic and corneal topographic data.
• Training and validation of models using 80% and 20% of the data, respectively.
• Comparison of ML-derived first alignment curve (AC1) with formula calculations.
• Application of multiple regression models (SVM, Gaussian process, decision tree, random forest) for axial length prediction.

Main Results:

• The ML model for OK lens fitting demonstrated superior performance (R²=0.93, MAE=0.19, MSE=0.09) compared to formula calculation (R²=0.66).
• High predictive accuracy (R² ranging from 0.91 to 0.96) was maintained across different OK lens brands and designs.
• The ML model for axial length prediction achieved a high R² value of 0.94, indicating excellent accuracy and robustness.

Conclusions:

• ML models significantly enhance the accuracy and efficiency of orthokeratology lens fitting.
• These models provide robust predictions for axial length changes after one year of OK lens wear.
• The developed models serve as valuable tools for guiding clinical decisions in orthokeratology practice.